1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2002 Cluster File Systems, Inc.
6 * This file is part of the Lustre file system, http://www.lustre.org
7 * Lustre is a trademark of Cluster File Systems, Inc.
9 * You may have signed or agreed to another license before downloading
10 * this software. If so, you are bound by the terms and conditions
11 * of that agreement, and the following does not apply to you. See the
12 * LICENSE file included with this distribution for more information.
14 * If you did not agree to a different license, then this copy of Lustre
15 * is open source software; you can redistribute it and/or modify it
16 * under the terms of version 2 of the GNU General Public License as
17 * published by the Free Software Foundation.
19 * In either case, Lustre is distributed in the hope that it will be
20 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
21 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
22 * license text for more details.
25 #define DEBUG_SUBSYSTEM S_CLASS
28 # include <liblustre.h>
31 #include <obd_support.h>
33 #include <lprocfs_status.h>
34 #include <lustre/lustre_idl.h>
35 #include <lustre_net.h>
36 #include <obd_class.h>
37 #include "ptlrpc_internal.h"
40 struct ll_rpc_opcode {
43 } ll_rpc_opcode_table[LUSTRE_MAX_OPCODES] = {
44 { OST_REPLY, "ost_reply" },
45 { OST_GETATTR, "ost_getattr" },
46 { OST_SETATTR, "ost_setattr" },
47 { OST_READ, "ost_read" },
48 { OST_WRITE, "ost_write" },
49 { OST_CREATE , "ost_create" },
50 { OST_DESTROY, "ost_destroy" },
51 { OST_GET_INFO, "ost_get_info" },
52 { OST_CONNECT, "ost_connect" },
53 { OST_DISCONNECT, "ost_disconnect" },
54 { OST_PUNCH, "ost_punch" },
55 { OST_OPEN, "ost_open" },
56 { OST_CLOSE, "ost_close" },
57 { OST_STATFS, "ost_statfs" },
58 { 14, NULL }, /* formerly OST_SAN_READ */
59 { 15, NULL }, /* formerly OST_SAN_WRITE */
60 { OST_SYNC, "ost_sync" },
61 { OST_SET_INFO, "ost_set_info" },
62 { OST_QUOTACHECK, "ost_quotacheck" },
63 { OST_QUOTACTL, "ost_quotactl" },
64 { MDS_GETATTR, "mds_getattr" },
65 { MDS_GETATTR_NAME, "mds_getattr_lock" },
66 { MDS_CLOSE, "mds_close" },
67 { MDS_REINT, "mds_reint" },
68 { MDS_READPAGE, "mds_readpage" },
69 { MDS_CONNECT, "mds_connect" },
70 { MDS_DISCONNECT, "mds_disconnect" },
71 { MDS_GETSTATUS, "mds_getstatus" },
72 { MDS_STATFS, "mds_statfs" },
73 { MDS_PIN, "mds_pin" },
74 { MDS_UNPIN, "mds_unpin" },
75 { MDS_SYNC, "mds_sync" },
76 { MDS_DONE_WRITING, "mds_done_writing" },
77 { MDS_SET_INFO, "mds_set_info" },
78 { MDS_QUOTACHECK, "mds_quotacheck" },
79 { MDS_QUOTACTL, "mds_quotactl" },
80 { MDS_GETXATTR, "mds_getxattr" },
81 { MDS_SETXATTR, "mds_setxattr" },
82 { MDS_WRITEPAGE, "mds_writepage" },
83 { MDS_IS_SUBDIR, "mds_is_subdir" },
84 { LDLM_ENQUEUE, "ldlm_enqueue" },
85 { LDLM_CONVERT, "ldlm_convert" },
86 { LDLM_CANCEL, "ldlm_cancel" },
87 { LDLM_BL_CALLBACK, "ldlm_bl_callback" },
88 { LDLM_CP_CALLBACK, "ldlm_cp_callback" },
89 { LDLM_GL_CALLBACK, "ldlm_gl_callback" },
90 { MGS_CONNECT, "mgs_connect" },
91 { MGS_DISCONNECT, "mgs_disconnect" },
92 { MGS_EXCEPTION, "mgs_exception" },
93 { MGS_TARGET_REG, "mgs_target_reg" },
94 { MGS_TARGET_DEL, "mgs_target_del" },
95 { MGS_SET_INFO, "mgs_set_info" },
96 { OBD_PING, "obd_ping" },
97 { OBD_LOG_CANCEL, "llog_origin_handle_cancel" },
98 { OBD_QC_CALLBACK, "obd_qc_callback" },
99 { LLOG_ORIGIN_HANDLE_CREATE, "llog_origin_handle_create" },
100 { LLOG_ORIGIN_HANDLE_NEXT_BLOCK, "llog_origin_handle_next_block" },
101 { LLOG_ORIGIN_HANDLE_READ_HEADER,"llog_origin_handle_read_header" },
102 { LLOG_ORIGIN_HANDLE_WRITE_REC, "llog_origin_handle_write_rec" },
103 { LLOG_ORIGIN_HANDLE_CLOSE, "llog_origin_handle_close" },
104 { LLOG_ORIGIN_CONNECT, "llog_origin_connect" },
105 { LLOG_CATINFO, "llog_catinfo" },
106 { LLOG_ORIGIN_HANDLE_PREV_BLOCK, "llog_origin_handle_prev_block" },
107 { LLOG_ORIGIN_HANDLE_DESTROY, "llog_origin_handle_destroy" },
108 { FLD_QUERY, "fld_query" },
109 { SEQ_QUERY, "seq_query" },
110 { SEC_CTX_INIT, "sec_ctx_init" },
111 { SEC_CTX_INIT_CONT,"sec_ctx_init_cont" },
112 { SEC_CTX_FINI, "sec_ctx_fini" }
118 } ll_eopcode_table[EXTRA_LAST_OPC] = {
119 { LDLM_GLIMPSE_ENQUEUE, "ldlm_glimpse_enqueue" },
120 { LDLM_PLAIN_ENQUEUE, "ldlm_plain_enqueue" },
121 { LDLM_EXTENT_ENQUEUE, "ldlm_extent_enqueue" },
122 { LDLM_FLOCK_ENQUEUE, "ldlm_flock_enqueue" },
123 { LDLM_IBITS_ENQUEUE, "ldlm_ibits_enqueue" },
124 { MDS_REINT_CREATE, "mds_reint_create" },
125 { MDS_REINT_LINK, "mds_reint_link" },
126 { MDS_REINT_OPEN, "mds_reint_open" },
127 { MDS_REINT_SETATTR, "mds_reint_setattr" },
128 { MDS_REINT_RENAME, "mds_reint_rename" },
129 { MDS_REINT_UNLINK, "mds_reint_unlink" }
132 const char *ll_opcode2str(__u32 opcode)
134 /* When one of the assertions below fail, chances are that:
135 * 1) A new opcode was added in lustre_idl.h, but was
136 * is missing from the table above.
137 * or 2) The opcode space was renumbered or rearranged,
138 * and the opcode_offset() function in
139 * ptlrpc_internal.h needs to be modified.
141 __u32 offset = opcode_offset(opcode);
142 LASSERT(offset < LUSTRE_MAX_OPCODES);
143 LASSERT(ll_rpc_opcode_table[offset].opcode == opcode);
144 return ll_rpc_opcode_table[offset].opname;
147 const char* ll_eopcode2str(__u32 opcode)
149 LASSERT(ll_eopcode_table[opcode].opcode == opcode);
150 return ll_eopcode_table[opcode].opname;
153 void ptlrpc_lprocfs_register(struct proc_dir_entry *root, char *dir,
154 char *name, struct proc_dir_entry **procroot_ret,
155 struct lprocfs_stats **stats_ret)
157 struct proc_dir_entry *svc_procroot;
158 struct lprocfs_stats *svc_stats;
160 unsigned int svc_counter_config = LPROCFS_CNTR_AVGMINMAX |
163 LASSERT(*procroot_ret == NULL);
164 LASSERT(*stats_ret == NULL);
166 svc_stats = lprocfs_alloc_stats(EXTRA_MAX_OPCODES + LUSTRE_MAX_OPCODES, 0);
167 if (svc_stats == NULL)
171 svc_procroot = lprocfs_register(dir, root, NULL, NULL);
172 if (IS_ERR(svc_procroot)) {
173 lprocfs_free_stats(&svc_stats);
180 lprocfs_counter_init(svc_stats, PTLRPC_REQWAIT_CNTR,
181 svc_counter_config, "req_waittime", "usec");
182 lprocfs_counter_init(svc_stats, PTLRPC_REQQDEPTH_CNTR,
183 svc_counter_config, "req_qdepth", "reqs");
184 lprocfs_counter_init(svc_stats, PTLRPC_REQACTIVE_CNTR,
185 svc_counter_config, "req_active", "reqs");
186 lprocfs_counter_init(svc_stats, PTLRPC_TIMEOUT,
187 svc_counter_config, "req_timeout", "sec");
188 lprocfs_counter_init(svc_stats, PTLRPC_REQBUF_AVAIL_CNTR,
189 svc_counter_config, "reqbuf_avail", "bufs");
190 for (i = 0; i < EXTRA_LAST_OPC; i++) {
191 lprocfs_counter_init(svc_stats, PTLRPC_LAST_CNTR + i,
193 ll_eopcode2str(i), "reqs");
195 for (i = 0; i < LUSTRE_MAX_OPCODES; i++) {
196 __u32 opcode = ll_rpc_opcode_table[i].opcode;
197 lprocfs_counter_init(svc_stats,
198 EXTRA_MAX_OPCODES + i, svc_counter_config,
199 ll_opcode2str(opcode), "usec");
202 rc = lprocfs_register_stats(svc_procroot, name, svc_stats);
205 lprocfs_remove(&svc_procroot);
206 lprocfs_free_stats(&svc_stats);
209 *procroot_ret = svc_procroot;
210 *stats_ret = svc_stats;
215 ptlrpc_lprocfs_read_req_history_len(char *page, char **start, off_t off,
216 int count, int *eof, void *data)
218 struct ptlrpc_service *svc = data;
221 return snprintf(page, count, "%d\n", svc->srv_n_history_rqbds);
225 ptlrpc_lprocfs_read_req_history_max(char *page, char **start, off_t off,
226 int count, int *eof, void *data)
228 struct ptlrpc_service *svc = data;
231 return snprintf(page, count, "%d\n", svc->srv_max_history_rqbds);
235 ptlrpc_lprocfs_write_req_history_max(struct file *file, const char *buffer,
236 unsigned long count, void *data)
238 struct ptlrpc_service *svc = data;
241 int rc = lprocfs_write_helper(buffer, count, &val);
249 /* This sanity check is more of an insanity check; we can still
250 * hose a kernel by allowing the request history to grow too
252 bufpages = (svc->srv_buf_size + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
253 if (val > num_physpages/(2 * bufpages))
256 spin_lock(&svc->srv_lock);
257 svc->srv_max_history_rqbds = val;
258 spin_unlock(&svc->srv_lock);
263 struct ptlrpc_srh_iterator {
265 struct ptlrpc_request *srhi_req;
269 ptlrpc_lprocfs_svc_req_history_seek(struct ptlrpc_service *svc,
270 struct ptlrpc_srh_iterator *srhi,
274 struct ptlrpc_request *req;
276 if (srhi->srhi_req != NULL &&
277 srhi->srhi_seq > svc->srv_request_max_cull_seq &&
278 srhi->srhi_seq <= seq) {
279 /* If srhi_req was set previously, hasn't been culled and
280 * we're searching for a seq on or after it (i.e. more
281 * recent), search from it onwards.
282 * Since the service history is LRU (i.e. culled reqs will
283 * be near the head), we shouldn't have to do long
285 LASSERT (srhi->srhi_seq == srhi->srhi_req->rq_history_seq);
286 LASSERT (!list_empty(&svc->srv_request_history));
287 e = &srhi->srhi_req->rq_history_list;
289 /* search from start */
290 e = svc->srv_request_history.next;
293 while (e != &svc->srv_request_history) {
294 req = list_entry(e, struct ptlrpc_request, rq_history_list);
296 if (req->rq_history_seq >= seq) {
297 srhi->srhi_seq = req->rq_history_seq;
298 srhi->srhi_req = req;
308 ptlrpc_lprocfs_svc_req_history_start(struct seq_file *s, loff_t *pos)
310 struct ptlrpc_service *svc = s->private;
311 struct ptlrpc_srh_iterator *srhi;
314 OBD_ALLOC(srhi, sizeof(*srhi));
319 srhi->srhi_req = NULL;
321 spin_lock(&svc->srv_lock);
322 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos);
323 spin_unlock(&svc->srv_lock);
326 *pos = srhi->srhi_seq;
330 OBD_FREE(srhi, sizeof(*srhi));
335 ptlrpc_lprocfs_svc_req_history_stop(struct seq_file *s, void *iter)
337 struct ptlrpc_srh_iterator *srhi = iter;
340 OBD_FREE(srhi, sizeof(*srhi));
344 ptlrpc_lprocfs_svc_req_history_next(struct seq_file *s,
345 void *iter, loff_t *pos)
347 struct ptlrpc_service *svc = s->private;
348 struct ptlrpc_srh_iterator *srhi = iter;
351 spin_lock(&svc->srv_lock);
352 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, *pos + 1);
353 spin_unlock(&svc->srv_lock);
356 OBD_FREE(srhi, sizeof(*srhi));
360 *pos = srhi->srhi_seq;
364 /* common ost/mdt srv_request_history_print_fn */
365 void target_print_req(void *seq_file, struct ptlrpc_request *req)
367 /* Called holding srv_lock with irqs disabled.
368 * Print specific req contents and a newline.
369 * CAVEAT EMPTOR: check request message length before printing!!!
370 * You might have received any old crap so you must be just as
371 * careful here as the service's request parser!!! */
372 struct seq_file *sf = seq_file;
374 switch (req->rq_phase) {
376 /* still awaiting a service thread's attention, or rejected
377 * because the generic request message didn't unpack */
378 seq_printf(sf, "<not swabbed>\n");
380 case RQ_PHASE_INTERPRET:
381 /* being handled, so basic msg swabbed, and opc is valid
382 * but racing with mds_handle() */
383 case RQ_PHASE_COMPLETE:
384 /* been handled by mds_handle() reply state possibly still
386 seq_printf(sf, "opc %d\n", lustre_msg_get_opc(req->rq_reqmsg));
389 DEBUG_REQ(D_ERROR, req, "bad phase %d", req->rq_phase);
392 EXPORT_SYMBOL(target_print_req);
394 static int ptlrpc_lprocfs_svc_req_history_show(struct seq_file *s, void *iter)
396 struct ptlrpc_service *svc = s->private;
397 struct ptlrpc_srh_iterator *srhi = iter;
398 struct ptlrpc_request *req;
401 spin_lock(&svc->srv_lock);
403 rc = ptlrpc_lprocfs_svc_req_history_seek(svc, srhi, srhi->srhi_seq);
406 req = srhi->srhi_req;
408 /* Print common req fields.
409 * CAVEAT EMPTOR: we're racing with the service handler
410 * here. The request could contain any old crap, so you
411 * must be just as careful as the service's request
412 * parser. Currently I only print stuff here I know is OK
413 * to look at coz it was set up in request_in_callback()!!! */
414 seq_printf(s, LPD64":%s:%s:x"LPD64":%d:%s:%ld:%lds(%+lds) ",
415 req->rq_history_seq, libcfs_nid2str(req->rq_self),
416 libcfs_id2str(req->rq_peer), req->rq_xid,
417 req->rq_reqlen, ptlrpc_rqphase2str(req),
418 req->rq_arrival_time.tv_sec,
419 req->rq_sent - req->rq_arrival_time.tv_sec,
420 req->rq_sent - req->rq_deadline);
421 if (svc->srv_request_history_print_fn == NULL)
424 svc->srv_request_history_print_fn(s, srhi->srhi_req);
427 spin_unlock(&svc->srv_lock);
433 ptlrpc_lprocfs_svc_req_history_open(struct inode *inode, struct file *file)
435 static struct seq_operations sops = {
436 .start = ptlrpc_lprocfs_svc_req_history_start,
437 .stop = ptlrpc_lprocfs_svc_req_history_stop,
438 .next = ptlrpc_lprocfs_svc_req_history_next,
439 .show = ptlrpc_lprocfs_svc_req_history_show,
441 struct proc_dir_entry *dp = PDE(inode);
442 struct seq_file *seqf;
445 LPROCFS_ENTRY_AND_CHECK(dp);
446 rc = seq_open(file, &sops);
452 seqf = file->private_data;
453 seqf->private = dp->data;
457 /* See also lprocfs_rd_timeouts */
458 static int ptlrpc_lprocfs_rd_timeouts(char *page, char **start, off_t off,
459 int count, int *eof, void *data)
461 struct ptlrpc_service *svc = data;
462 unsigned int cur, worst;
468 cur = at_get(&svc->srv_at_estimate);
469 worst = svc->srv_at_estimate.at_worst_ever;
470 worstt = svc->srv_at_estimate.at_worst_time;
471 s2dhms(&ts, cfs_time_current_sec() - worstt);
473 rc += snprintf(page + rc, count - rc,
474 "adaptive timeouts off, using obd_timeout %u\n",
476 rc += snprintf(page + rc, count - rc,
477 "%10s : cur %3u worst %3u (at %ld, "DHMS_FMT" ago) ",
478 "service", cur, worst, worstt,
480 rc = lprocfs_at_hist_helper(page, count, rc,
481 &svc->srv_at_estimate);
485 void ptlrpc_lprocfs_register_service(struct proc_dir_entry *entry,
486 struct ptlrpc_service *svc)
488 struct lprocfs_vars lproc_vars[] = {
489 {.name = "req_buffer_history_len",
491 .read_fptr = ptlrpc_lprocfs_read_req_history_len,
493 {.name = "req_buffer_history_max",
494 .write_fptr = ptlrpc_lprocfs_write_req_history_max,
495 .read_fptr = ptlrpc_lprocfs_read_req_history_max,
498 .read_fptr = ptlrpc_lprocfs_rd_timeouts,
502 static struct file_operations req_history_fops = {
503 .owner = THIS_MODULE,
504 .open = ptlrpc_lprocfs_svc_req_history_open,
507 .release = lprocfs_seq_release,
512 ptlrpc_lprocfs_register(entry, svc->srv_name,
513 "stats", &svc->srv_procroot,
516 if (svc->srv_procroot == NULL)
519 lprocfs_add_vars(svc->srv_procroot, lproc_vars, NULL);
521 rc = lprocfs_seq_create(svc->srv_procroot, "req_history",
522 0400, &req_history_fops, svc);
524 CWARN("Error adding the req_history file\n");
527 void ptlrpc_lprocfs_register_obd(struct obd_device *obddev)
529 ptlrpc_lprocfs_register(obddev->obd_proc_entry, NULL, "stats",
530 &obddev->obd_svc_procroot,
531 &obddev->obd_svc_stats);
533 EXPORT_SYMBOL(ptlrpc_lprocfs_register_obd);
535 void ptlrpc_lprocfs_rpc_sent(struct ptlrpc_request *req)
537 struct lprocfs_stats *svc_stats;
538 __u32 op = lustre_msg_get_opc(req->rq_reqmsg);
539 int opc = opcode_offset(op);
541 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
542 if (svc_stats == NULL || opc <= 0)
544 LASSERT(opc < LUSTRE_MAX_OPCODES);
545 /* These two use the ptlrpc_lprocfs_brw below */
546 if (!(opc == OST_WRITE || opc == OST_READ || op == LDLM_ENQUEUE
548 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, 0);
551 void ptlrpc_lprocfs_brw(struct ptlrpc_request *req, int opc, int bytes)
553 struct lprocfs_stats *svc_stats;
554 svc_stats = req->rq_import->imp_obd->obd_svc_stats;
557 lprocfs_counter_add(svc_stats, opc + EXTRA_MAX_OPCODES, bytes);
559 EXPORT_SYMBOL(ptlrpc_lprocfs_brw);
561 void ptlrpc_lprocfs_unregister_service(struct ptlrpc_service *svc)
563 if (svc->srv_procroot != NULL)
564 lprocfs_remove(&svc->srv_procroot);
567 lprocfs_free_stats(&svc->srv_stats);
570 void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd)
572 if (obd->obd_svc_procroot)
573 lprocfs_remove(&obd->obd_svc_procroot);
575 if (obd->obd_svc_stats)
576 lprocfs_free_stats(&obd->obd_svc_stats);
578 EXPORT_SYMBOL(ptlrpc_lprocfs_unregister_obd);
581 int lprocfs_wr_evict_client(struct file *file, const char *buffer,
582 unsigned long count, void *data)
584 struct obd_device *obd = data;
585 char tmpbuf[sizeof(struct obd_uuid)];
587 /* Kludge code(deadlock situation): the lprocfs lock has been held
588 * since the client is evicted by writting client's
589 * uuid/nid to procfs "evict_client" entry. However,
590 * obd_export_evict_by_uuid() will call lprocfs_remove() to destroy
591 * the proc entries under the being destroyed export{}, so I have
592 * to drop the lock at first here.
593 * - jay, jxiong@clusterfs.com */
597 sscanf(buffer, "%40s", tmpbuf);
598 if (strncmp(tmpbuf, "nid:", 4) == 0)
599 obd_export_evict_by_nid(obd, tmpbuf + 4);
600 else if (strncmp(tmpbuf, "uuid:", 5) == 0)
601 obd_export_evict_by_uuid(obd, tmpbuf + 5);
603 obd_export_evict_by_uuid(obd, tmpbuf);
610 EXPORT_SYMBOL(lprocfs_wr_evict_client);
612 int lprocfs_wr_ping(struct file *file, const char *buffer,
613 unsigned long count, void *data)
615 struct obd_device *obd = data;
616 struct ptlrpc_request *req;
620 LPROCFS_CLIMP_CHECK(obd);
621 req = ptlrpc_request_alloc_pack(obd->u.cli.cl_import, &RQF_OBD_PING,
622 LUSTRE_OBD_VERSION, OBD_PING);
624 LPROCFS_CLIMP_EXIT(obd);
628 ptlrpc_request_set_replen(req);
629 req->rq_send_state = LUSTRE_IMP_FULL;
630 req->rq_no_resend = 1;
631 req->rq_no_delay = 1;
633 rc = ptlrpc_queue_wait(req);
635 ptlrpc_req_finished(req);
640 EXPORT_SYMBOL(lprocfs_wr_ping);